Impact of features selected by Principal Component Analysis in featured based steganalysis in calibrated and non-calibrated images

Steganalysis is helpful in finding the hidden information/data/message in cover images. In simple form, the confidential and concealed message has to be extracted efficiently in steganalysis. This paper performs universal steganalysis based on the features using F5 and Pixel Value Differencing (PVD) algorithms. The feature extraction is carried out through Discrete Cosine Transform (DCT) techniques. The dimensions or size of the feature vector/ feature set is reasonably diminished by Principal Component Analysis (PCA). The extracted features are the combined DCT and Markovian features that have 274 features. These features are inputted to the Linear Support Vector Machine (SVM) for classifying the stego and cover image. Prior to analysis, the images are calibrated so as to improve the efficiency of classifier. The analysis is done with different embedding percentages and the training and testing images are split in the ratio of 80 and 20 for SVM classifier